fix(cli): exit the server when the MCP client closes stdin

[HarshCasper]

Problem

In Docker mode, MCP server containers accumulate on the host. Every client session starts one container, and none of them exit when the session ends. A user reported collecting eight running localstack/localstack-mcp-server containers in a day. The npx entry has the same defect and leaks host Node processes instead.

Cause

The stdio transport only subscribes to data and error on stdin. Closing stdin is the shutdown signal defined by the MCP spec for stdio servers, and Docker delivers that EOF reliably: when the client detaches, the daemon closes the container's stdin (StdinOnce). The server ignores it. Once any live handle exists in the process (telemetry flush timers, keep-alive sockets, log streams), the event loop never drains and the process runs forever. --rm never fires because it requires the container to exit.

Change

• src/cli/lifecycle.ts: exit the process when stdin emits end or close. A 1.25 s grace period lets the PostHog client flush the last telemetry batch (its flushInterval is 1 s).
• The CLI launcher registers the watcher before loading stdio.js, so the published bin picks it up for npx users.
• The Docker image now boots through dist/cli.js instead of dist/stdio.js, which gives containers the same behavior. docker stop keeps working through the SIGTERM handler that xmcp installs, and the image build now smoke-tests the new entrypoint.
• New troubleshooting row in docs/DOCKER.md with a one-liner to remove stray containers from older images.

[gtsiolis]

Fix looks correct, left one comment worth addressing. 🙏